|
Comparison and evaluation of machine learning methods for salinity prediction in the Karun River
|
Hany Mahbuby *   , Mohammad Pirayesh , Yahya Djamour |
| Shahid Beheshti University |
|
|
Abstract: (333 Views) |
Most regions of Iran are characterized as a hot and arid climate with low precipitation. The primary water resources in the country are groundwater and river water. In recent years, due to little rainfall and drought conditions, rivers have received more attention. In this context, river water salinity is considered as one of the most critical water quality parameters, requiring special attention.
In this study, the long-term variations in the salinity of the Karun River were assessed. Furthermore, the monthly average salinity of the Karun River was modeled from 2001 (1380) to 2016 (1395) using various machine learning methods, including deep neural networks (DNN), random forests (RF), and extreme gradient boosting (XGBoost). Subsequently, the average salinity was forecasted for an 18-month period extending to mid-2018 (1397). The predicted values were compared with the observed measurements during this 18-month validation period, and the performance and accuracy of the different modeling approaches were evaluated and compared.
The results demonstrated that, first, the monthly average salinity increased at a rate of approximately 10 ppm/year throughout the study period, posing a potential threat to the regional ecosystem. The DNN and RF models showed comparable accuracy and could predict the average salinity of the river with the accuracy of 170 _ 180 ppm when their respective hyperparameters were optimally tuned. However, the RF performed slightly better in long-term forecasting.
Among the tested methods, XGBoost outperformed the others, achieving a prediction error of approximately 150 ppm. Compared to RF and DNN, this represents a relative error reduction of about 13% and 18%, respectively.
|
|
| Keywords: River salinity, Deep neural network, Random forest, Extreme gradient boosting |
|
|
Full-Text [PDF 1929 kb]
(16 Downloads)
|
Type of Study: Research |
Subject:
Hydrography
Received: 2025/05/11 | Accepted: 2025/07/29 | ePublished ahead of print: 2025/08/5 | Published: 2025/08/31
|
|
|
|
|
|
|
| Send email to the article author |
|
|
